Tensorboard estimator export_savedmodel简单用法

# add export graph
def serving_input_fn():
    input_ids = tf.placeholder(dtype=tf.int32, shape=[None, FLAGS.max_seq_length], name="input_ids")
    input_mask = tf.placeholder(dtype=tf.int32, shape=[None, FLAGS.max_seq_length], name="input_mask")
    segment_ids = tf.placeholder(dtype=tf.int32, shape=[None, FLAGS.max_seq_length], name="segment_ids")
    label_ids = tf.placeholder(dtype=tf.int32, shape=[None], name="label_ids")
    string_2 = tf.placeholder(dtype=tf.string, shape=[None], name="string_2")

    features = {'input_ids': input_ids,
                'input_mask': input_mask,
                'segment_ids': segment_ids,
                'label_ids': label_ids,
                'string_2': string_2}
    return tf.contrib.learn.InputFnOps(features, None, default_inputs=features)

export_dir = os.path.join(FLAGS.output_dir, "saved_model")
estimator.export_savedmodel(export_dir, serving_input_fn(),
                            assets_extra={"vocab.txt": FLAGS.vocab_file},
                            as_text=False,
                            strip_default_attrs=True)